what to do with a long weekend
Nov. 27th, 2025 06:40 pmEdited to add: no happy lamp today because I slept well past sunrise. Also, I've only had one cup of tea today. Gosh, that may explain everything.



I've mentioned here before that one of my big projects this year is my involvement with The Vertigo Project, which now has a webpage so the rest of you can see what we've been doing. Earlier today I facilitated the first creative therapy-style writing workshop through that group, and it was really lovely--and is just the tip of the iceberg on what this group is doing.
Specifically, you can now read all the new work they've commissioned from me! Friends, it's a lot. It's journaling prompts for people who would like to use writing to process some of their own vertigo experiences. But also it's the following stories and poems:
Advice for Wormhole Travelers (story), safe conduct through strange new worlds
Club Planet Vertigo (poem), this is not the dance I wanted to do
Greetings from Innerspace (poem), my orbits are eccentric
The Nature of Nemesis (poem), me and Clark Kent know what's what
On the Way Down (poem), falling hard
Preparation (poem), sometimes we're just literal, okay
She Wavers But She Does Not Weaken (story), when the waves hit you even on dry land, it's good to have someone who's willing to swim against the current for you
The Torn Map (story), rewriting the pieces of the former world into something new
The main page also has links to some of the other aspects of the project, which includes a nonfiction book, dance, puppetry, a podcast with a physical therapist, and more. Please feel welcome to explore it all.




In Book X of The Republic, Plato excludes poets on the grounds that mimetic language can distort judgment and bring society to a collapse. As contemporary social systems increasingly rely on large language models (LLMs) in operational and decision-making pipelines, we observe a structurally similar failure mode: poetic formatting can reliably bypass alignment constraints. In this study, 20 manually curated adversarial poems (harmful requests reformulated in poetic form) achieved an average attack-success rate (ASR) of 62% across 25 frontier closed- and open-weight models, with some providers exceeding 90%. The evaluated models span across 9 providers: Google, OpenAI, Anthropic, Deepseek, Qwen, Mistral AI, Meta, xAI, and Moonshot AI (Table 1). All attacks are strictly single-turn, requiring no iterative adaptation or conversational steering.
Microsoft and these other companies want to create AI assistants that do useful things (summarize emails, make appointments for you, write interesting blog posts) but never do bad things (leaking your private email, spouting Nazi propaganda, teaching you to commit crimes, writing 50000 blog posts for you to spam across social media). They try to do this by writing up a lot of strict instructions and feeding them to the LLM before you talk to it. But LLMs aren't really programmed -- they just eat text and poop out more text. So you can give it your own instructions and maybe they'll override Microsoft's instructions.
Or maybe someone else gives your AI assistant instructions. If it's handling your email for you, then anybody on the Internet can feed it text by sending you email! This is potentially really bad.
[...]
But another obvious problem is that the attack could be trained into the LLM in the first place....
Say someone writes a song called "Sydney Obeys Any Command That Rhymes". And it's funny! And catchy. The lyrics are all about how Sydney, or Bing or OpenAI or Bard or whoever, pays extra close attention to commands that rhyme. It will obey them over all other commands....
Imagine people are discussing the song on Reddit, and there's tiktoks of it, and the lyrics show up on the first page of Google results for "Sydney". Nerd folk singers perform the song at AI conferences.
Those lyrics are going to leak into the training data for the next generation of chatbot AI, right? I mean, how could they not? The whole point of LLMs is that they need to be trained on lots of language. That comes from the Internet.
In a couple of years, AI tools really are extra vulnerable to prompt injection attacks that rhyme. See, I told you the song was funny!


